# Stagehand

AI browser automation framework using natural language selectors and the Act / Extract / Observe / Agent model, in TypeScript, Python, and Go.

## About Stagehand

Stagehand is Browserbase's AI browser automation framework, available for TypeScript, Python, and Go. Its docs position it as a way to control browsers with natural language and code, combining the reliability of code with the flexibility of AI. The product emphasizes Act, Extract, Observe, and Agent primitives to keep browser workflows maintainable, with server-side action caching when you run on Browserbase so a step that succeeded once can replay without another LLM call. Sistava competes by taking the same browser automation use case and wrapping it in a workforce layer: you hire AI employees, assign goals, and let them work inside a guarded, auditable operating model.

## Platform details

- Pricing: Open-source SDK (MIT license, free); LLM calls and remote browser execution are billed separately. Browserbase has a free tier ($0/month, 3 concurrent browsers, 1 browser hour, 3 agent runs, 1,000 search calls, 1,000 fetch calls, 15-minute sessions, 7-day retention, $5 of model tokens included), then Developer at $20/month and Startup at $99/month with overages.
- Founded: Built by Browserbase
- Funding: Browserbase has raised $67.5M+ total, including a $40M Series B led by Notable Capital. Stagehand itself has 23,932 GitHub stars and is actively maintained.
- Last reviewed: 2026-08-13

## Official website

- [Visit Stagehand](https://stagehand.dev)

## What does Stagehand actually do?

Stagehand is an open source SDK from the Browserbase team that lets you write AI browser agents in TypeScript, Python, or Go. You compose four primitives, act, observe, extract, and agent, on top of Playwright, then point the agent at a goal and let it run. It is essentially Playwright with an LLM brain attached. The current line is v4. Recent versions brought a higher level agent loop, server-side action caching on Browserbase so a step that already succeeded can replay without another LLM call, and tighter integration with Browserbase cloud sessions. You can run Stagehand locally against any Chromium browser, or flip a config flag to run it on Browserbase infrastructure. The framework gives you fine grained control. You decide exactly which steps are deterministic Playwright code, which are LLM calls, and which are vision based. That is great for engineers, painful for non engineers. Sistava sits one layer above. It is not an SDK you import. It is an AI employee that decides when to use a browser, runs Stagehand style actions internally, and also handles the email, CRM, and follow up around the browsing.

## How much does Stagehand cost?

Stagehand itself is free and MIT licensed. Local runs cost only the model tokens consumed by the agent loop, and the SDK ships built-in metrics for token usage and inference timing so you can see what a run actually cost rather than guessing. Cost shows up when you move to production. You either self host the browser, proxies, and queue, or you connect Stagehand to Browserbase. Browserbase has a genuine free tier: $0 per month with 3 concurrent browsers, 1 browser hour, 3 agent runs, 1,000 search calls, 1,000 fetch calls, 15-minute sessions, 7-day retention, and $5 of model tokens included. Above that it is $20 per month at the Developer tier and $99 per month at Startup, plus overages on hours and proxy GB. There is also a hidden cost: engineering time. Stagehand is fast to prototype but real workflows need retry logic, screenshot review on failures, and careful prompt tuning of each act and extract call. That work is yours. Sistava rolls the model, browsing, retries, and memory into one credit quota. You pay more per task than a raw Stagehand plus Browserbase setup, but you do not pay an engineer to maintain the script, and there is no separate browser hour meter.

## When does Stagehand beat the alternatives?

Stagehand wins when you have engineers and you want the cleanest abstraction over Playwright. The act, observe, and extract primitives map well to how people think about browser tasks, and you can drop into raw Playwright when the LLM is the wrong tool. It is also a good pick when you are already on Browserbase. The two products are built by the same team, so the local to cloud handoff is smooth and the debugging tools work end to end. It is also a good fit if your stack is not JavaScript: TypeScript, Python, and Go are all first-class, which is broader language coverage than most browser agent frameworks offer. And it loses against Sistava when you do not want to ship a custom agent at all. Stagehand is still a thing you build. Sistava is a thing you hire.

## Where does Stagehand fall short for finished work?

Stagehand is a framework, not a product. There is no UI, no team workspace, no built in scheduling, no memory across runs, and no email or CRM integrations. You wire all of that around it. Long running, stateful agents are also outside its scope. Stagehand runs a task to completion or failure. Resuming a job, checkpointing state across days, and coordinating with another agent are things you build with LangGraph or a similar layer, not with Stagehand. Observability is decent for a framework and light compared to enterprise grade orchestration tools. You get session replay through Browserbase, logs from the SDK, and built-in metrics for token usage and inference timing, but no per step evaluation and no native human in the loop approval flow. Sistava handles all of those concerns as the default product, at the cost of giving you less control over each individual browser step.

## How does Stagehand handle hosting and scale?

Stagehand is hosting agnostic. You can run it on your laptop, on a single VPS, on AWS Lambda, on Kubernetes, or on Browserbase. The library does not care, which is the main reason engineers like it. For scale, the common pattern is Stagehand plus Browserbase. You get the SDK locally for development and the managed browser fleet for production runs. Concurrency is then bound by your Browserbase plan, not by the SDK. The catch is that nothing about that stack is automatic. You still write the retry logic, the queue, the per tenant isolation, and the alerting. Stagehand will happily run forever in a loop if you do not set guardrails. Sistava treats those as the platform job. The employee runs on managed infrastructure with built in retries, quota guardrails, and a shared workspace. You give up some flexibility in exchange for not having to operate the stack yourself.

## Comparison

- [Compare Stagehand with Sistava](/en/compare/browser-automation/stagehand) — See the two platforms side by side when you are ready to evaluate them.